KR20170134324A - METHODS FOR ENCODING DIGITAL IMAGES AND RELATED DECODING METHODS, DEVICES, AND COMPUTER PROGRAMS - Google Patents

Abstract

A method of encoding a digital image is disclosed, wherein the image (lk) is divided into a plurality of blocks of pixels processed in a predetermined order, the method comprising the following steps implemented for a current block (x) of predetermined sizes :
- transforming the current block into a transformed block (T2), the block comprising coefficients, the step implementing two consecutive transforming sub-steps, the first sub- Wherein the sub-step is applied to an intermediate block resulting from the first sub-step, the intermediate block comprising coefficients, and
- quantizing (T4) and encoding (T5) the coefficients of the transformed block,
Wherein the transforming step is performed prior to at least one transforming sub-step of transforming sub-steps, wherein the sub-step is applied to a block that is a so-called transformed block of a current block and an intermediate block, Further comprising spare sub-steps (T20, T23) for forming a first and at least one second distinct vector of blocks to be transformed, said vector having a length equal to one of the sizes of blocks to be transformed, Neighboring coefficients, individually the coefficients of the sequence of pixels, individually including pixels; And
- the at least one transform sub-step comprises applying a first transform (L0, C0) to the at least one first vector and applying at least one second transform to the at least one second vector of the block And applying transforms (L1, C1).

Description

METHODS FOR ENCODING DIGITAL IMAGES AND RELATED DECODING METHODS, DEVICES, AND COMPUTER PROGRAMS

The field of the invention is in the field of signal compression, and in particular the field of digital images or sequences of digital images, which are divided into blocks of pixels.

The encoding / decoding of digital images is particularly applied to images from at least one video sequence comprising:

- images from the same camera and temporally following each other (type-2D encoding / decoding),

- images (3D-type encoding / decoding) from different cameras oriented according to different views,

 - corresponding texture and depth components (3D-type encoding / decoding),

- Etc.

The present invention is applied in a similar manner to 2D-type or 3D-type encoding / decoding of images.

In particular, the present invention relates to a method and apparatus for video encoding that is currently implemented in Advanced Video Coding (AVC) and High Efficiency Video Coding (HEVC) video encoders and their extensions (MVC, 3D-AVC, MV-HEVC, 3D- And may be applied (but not exclusively) to the corresponding decoding.

Conventional compression schemes for digital images are contemplated, in which the image is divided into blocks of pixels. In general, the current block to be coded constituting the initial coding unit is cut into a variable number of sub-blocks according to a predetermined cutting mode. With reference to Fig. 1, the present inventors consider a sequence of digital images (I1, I1, Ik, IK) (K is a non-zero integer). CTUs according to the HEVC terminology specified in the document "ISO / IEC 23008-2: 2013 - High efficiency coding and media delivery in heterogeneous environments - Part 2: High efficiency video coding» (International Organization for Standardization, published in November 2013) The standard encoders typically provide regular partitioning, which is known as fixed-size CUs (in the case of "Coding Units"). The partitioning is always performed from the initial non-partitioned encoding unit, the final partitioning is computed, and thereafter is signaled from this neutral basis. The partitioning allowed by the HEVC standard Examples are presented in connection with FIG.

Each CU will undergo encoding or decoding operations consisting of a sequence of operations including prediction, remainder calculation, transform, quantization and entropic coding in a non-exclusive manner. The sequence of such operations is known from the prior art and is presented in connection with FIG.

The first block CTU to be processed is selected as the current block b. For example, this is the first block (in lexicographic order). This block contains NxN pixels, where N is a non-zero integer, e.g., equal to 64 according to the HEVC standard.

There are L partitionings to possible PU blocks numbered from 1 to L, and it is assumed that the partitioning used for block (b) corresponds to partitioning number (1). For example, there may be four possible partitions with sub-blocks of size 4x4, 8x8, 16x16, and 32x32, depending on the regular mode of "quadtree" -type cutting. As noted previously, some PU blocks may be rectangular.

During the step E1, the prediction Pr of the original block b is determined. It may be a block or block that is constructed by known means, typically by motion compensation (a block originating from a previously decoded reference image) or by intra prediction (a block composed of decoded pixels immediately adjacent to a current block in an ID image) Prediction block. The prediction information related to P is encoded into a bit stream (TB) or a compressed file (FC). Here, it is assumed that there are K possible prediction modes (m1, m2, ..., mK) (K is a non-zero integer). For example, the prediction mode selected for current block b is mode mk. Some prediction modes are associated with intra-type prediction, other prediction modes are associated with INTER-type prediction, and other prediction modes are associated with MERGE-type prediction.

During step E2, the original remainder R is formed by subtracting R = b-Pr from the prediction Pr of the current block b in the current block b.

During step E3, the remainder (R) is transformed into DCT-type transforms, or wavelet transforms, into transformed residual blocks called RT, both of which are known to those skilled in the art, And JPEG / MPEG standards for JPEG2000 for DCT and wavelet transforms.

In E4, the transformed residual (RT) is quantized into a residual quantized block (RQ) by a conventional quantization means such as a scalar or a vector, and the quantized residual block (RQ) (Nb is a non-zero integer) of the number of included pixels, e.g., Nb. In a manner known in the state of the art, these coefficients are scanned in a predetermined order to construct a one-dimensional vector RQ [i] (index (i) varies from 0 to Nb-1). The index (i) is called the frequency of the coefficient RQ [i]. Typically, these coefficients are scanned in ascending order of frequency, e.g., according to a zigzag path known from the JPEG fixed image encoding standard.

During step E5, the amplitude information of the coefficients of the remaining block RQ is encoded by entropy coding, for example, according to the Huffman encoding technique or the arithmetic encoding technique. Here, by amplitude, is regarded as the absolute value of the coefficient. Typically, it is possible to encode information representative of the fact that the coefficients are non-zero for each coefficient. Thereafter, for each non-zero coefficient, one or more pieces of information related to the amplitude are encoded. The encoded amplitudes CA are obtained. In general, the signs of the non-zero coefficients are simply coded to bit 0 or 1, and each value corresponds to a given polarity. Such encoding provides efficient performances because, due to the transform, the values of the amplitudes to be encoded are substantially equal to zero.

The preceding steps E1 to E5 are repeated for a possible 1 number of partitions of the current block b. In the case of partitioning 1, each of the sub-blocks CU of the current block b is processed as described above, and a prediction type (inter or intra) is granted for each CU. In other words, all the sub-blocks PU of the sub-block CU undergo the same prediction type.

For example, the coded data for each possible partitioning I is competing according to a rate-distortion criterion, and the partitioning to obtain the best results according to these criteria is finally adopted.

The other blocks of image I1 are processed in the same manner as for the subsequent images of the sequence.

The transformation step plays a critical role in this video coding scheme, in fact, it focuses the information before the quantization operation. Thus, the set of residual pixels before encoding is represented by a small number of transformed coefficients, also referred to as frequencies of non-zero frequency, representing the same information.

Thus, instead of transmitting a large number of coefficients, only a small number will be needed to reliably reconstruct the block of pixels.

The efficiency of the transform is generally measured in accordance with an energy concentration criterion also given in the form of a coding gain, which is a reduction in distortion when a coding is done in a transformed domain rather than a spatial domain, for a given bit rate ). ≪ / RTI >

는 변환된 도메인에서 수행된 양자화 이후의 평균 제곱 에러를 나타내고,

는 공간 도메인에서의 동일한 정밀도를 갖는 양자화에 대한 제곱 에러, 즉, 프로세싱 전의 잔여를 나타낸다.

Represents the mean squared error after quantization performed in the transformed domain,

Represents the squared error for the quantization with the same precision in the spatial domain, i. E., The residual before processing.

According to these notations, the following coding gain is expressed as: < RTI ID = 0.0 >

Generally, this gain is expressed in dB.

The gain realized in the distortion due to the use of the transforms can also be transcribed with the gain of the bit rate, at a high bit rate, the gain in dB divided by the value of 6.02 is the actual bit rate economical / RTI >

In image and video encoding, the most commonly used transforms are linear, orthogonal, or quasi-orthogonal transforms (4x4, 8x8, etc.). The linear transformation can be represented as a matrix as follows. The orthogonal transformation has characteristic features that the inverse transformation matrix is a transposition of the direct transformation matrix. Thus, this transformation has the following characteristics:

Where A is a matrix with direct transforms, I is an identity matrix, and c is a numerical value. When c is 1, the matrix A is normal orthogonal. The quasi-orthogonal matrix multiplied by its inverse has an approximate amount close to the identity matrix in one factor.

The most commonly used transforms are based on cosine bases. Thus, there is DCT in most image and video compression standards. Recently, the HEVC standard has also introduced DST (Discrete Sine Transform) for coding of certain remainders in the case of 4x4 size blocks.

In fact, approximations of these transforms are used, and calculations are performed with integers. In general, the transform bases are approximated to the nearest integer with a given precision (typically 8 bits).

By way of example, with reference to Figures 4A and 4B, transforms used by HEVC standard blocks of size 4x4 are presented.

These are DCT and DST transforms. The values presented in this table will be divided by 128 to restore quasi-normal orthogonal transformations.

In practice, the block of pixels is transformed by the following operations:

A represents a transform matrix of size N x N, x represents residual pixels or pixels to be transformed (spatial domain), X is a block of transformed domain (called frequency domain) It is an operator.

The blocks of the spatial domain are written as follows (in this case 4x4):

The block of pixels of the transformed domain takes the form:

Matrix A takes the form of a 4x4 matrix in this case, and the coefficients are the same as those shown in the tables of Figures 4a and 4b.

Therefore, the above equation is proceeded as follows.

· Block x is transposed.

· Thus, the matrix A is applied to the lines of pixels.

가 전치된다.· After that, the result

.

· The application of matrix A is then based on these results.

The combination of these operations will then convert the lines and then convert the columns.

From a coding standpoint, but equally mathematically, the different entries are made by doing the following:

In this case, the columns are first converted.

As described in the cited prior art, the transforms per line may be different. The above formulas can be extended to this case if, for example, the operator wishes to operate the lines after first operating the columns:

L is a line-specific conversion, and C is a column-specific conversion.

To pre-operate the lines, the following equation can be written: < RTI ID = 0.0 >

These last two formulas are also possible to process rectangular blocks of pixels, so that in the first step an 8x8 column transform can operate a 4x8 block, and in a second step, , A 4x4 line transform will operate the resulting block.

Thus, it is written in the same way as before:

It can be ascertained that two separate transformations are made to the line-column or to the line-column direction. Thus, the spatial signal is de-correlated by the line vectors, then the columns or columns, and then the lines.

Inverse correlation is in fact an important aspect of transformation. Ensuring good reverse correlation makes it possible to obtain good coding gain.

The conversion operations presented above are performed separately by line, then column inverse correlation or column, and then by line inverse correlation. These are so-called separable transformations.

As a result, the separable transforms do not allow complete decorrelation of pixels in the spatial domain, in particular decorrelation of diagonal elements such as x0 and x5.

In order to enable full downconversion, it is possible to perform a non-separable conversion. To do this, the inventors apply the following operators:

및

는 벡터들로 형성된 공간 도메인에서 각각 프로세싱된 픽셀들이다. 따라서, 4x4 블록의 경우,

및

는 아래의 형태를 취한다:A _ns denotes the non-separable processing to be performed and is advantageously chosen as normal orthonormal or quasi-normal orthonormal,

And

Are the pixels processed individually in the spatial domain formed of vectors. Therefore, in the case of a 4x4 block,

And

Takes the following form:

Thus, A is _ns in this example, has a size of 16x16, more commonly has a size of N ² xN ² for the NxN block size consisting of N ² pixels.

In this way, the non-separable transform A _ns can process all correlations between pixels in the spatial domain, including diagonal correlations, frontally. For example, in the case of a 4x4 block, the direct correlation between pixels x0 and x5 is reduced.

As a result, non-separable transforms are more efficient in compression, but non-separable transformations nonetheless have a certain number of disadvantages:

● non-separable, so the conversion to convert the size of N ² xN ^2, as compared with 2xN ^two values required for the two matrix for performing a separable transform, N ⁴ values have to be stored. This affects the cost of implementing an image / video encoder / decoder, which is expensive for high speed memories.

● The transformation algorithm requires a significantly larger number of operations to handle the increased size of the vector.

Thus, compared to separable transforms, the non-separable transformations require much higher complexity as shown in the table of FIG. In this table, operations and amounts of memory associated with implementations of non-separable (Nsep) and separable (Sep) transforms are listed for processing of 4x4 and 8x8 blocks, respectively.

As can be seen, non-separable transforms have a significant impact on the ROM and many operations as compared to separable transforms. In particular, in terms of operations, the complexity ratio is greater than M / 2 for a block of size MxM.

In addition, a patent application published under number US2013 / 0003828 selects the types of transformations to be applied successively to the columns and lines of this block based on the prediction mode selected for the block, e.g., the intra-directional prediction mode of the particular direction A method of encoding an image is disclosed. More precisely, the method associates a first linear transformation to the lines of the current block (e.g., DCT) and associates a second linear transformation, distinct from the first linear transformation, to the columns of the transformed block by the first transformation .

The advantage is to adapt to the fact that the set of lines and the set of columns of the block do not necessarily have the same statistical properties.

Disadvantages of the prior art are:

● For non-separable transformations, the algorithm complexity is very important. Therefore, it requires a very high or even substantial implementation cost.

● For separable transforms, separable transforms are performed identically for the lines or columns, and in particular, it is implicitly assumed that each column or set of lines each share the same statistics. Detachable transforms are the most commonly used transforms in current encoders because separable transforms provide a trade-off between compression efficiency and encoding cost. However, separable transforms do not reach the compressive capabilities of non-separable transforms.

The present invention actually improves the situation.

The object of the present invention is, in particular, to overcome these shortcomings of the prior art.

More precisely, it is an object of the present invention to propose a solution that improves the compression performance of a digital image encoder, without requiring a significant increase in computation and memory resources.

These objects as well as other objects which will appear below are achieved by a method for encoding a digital image, the image being divided into a plurality of blocks of pixels to be processed in a defined order, the method comprising the steps of: The following steps being implemented for the current block having:

- Transforming a current block into a transformed block, the block comprising coefficients, the step implementing two consecutive transforming sub-steps, wherein the first sub-step is applied to the current block, Wherein the sub-step is applied to an intermediate block resulting from the first sub-step, the intermediate block comprising coefficients, and

- Quantizing and encoding the coefficients of the transformed block.

According to the present invention, before at least one of said transforming sub-steps, which is applied to a block to be transformed, between a current block and an intermediate block, the method comprises forming at least first and second distinct vectors in a block to be transformed Wherein the vector comprises a sequence of pixels of a length equal to one of the sizes of the block to be transformed, individually adjacent coefficients, and wherein the at least sub- Applying a transform and applying at least one second transform different from the first transform to the at least one second vector of the block.

According to the present invention, at least one of two consecutive steps of a detachable transformation applied to a current block is a line of the current block, having a size that is individually equal to the size of the column, and the sequence of neighboring elements of the current block Lt; RTI ID = 0.0 > of the < / RTI >

Thus, the present invention provides a completely novel and novel method of image coding by transforming the pixels of the current block in the spatial domain into coefficients in the frequency domain, providing different processing for the two vectors in the current block It follows a creative approach.

In contrast to the prior art based on the hypothesis of reduction of common statistics shared by lines or columns of the same block of an image to be coded, the present invention is based on the assumption that two vectors of the same size, separate from the block, Which may have different statistical properties.

The present invention presents the same algorithm complexity as the prior art, but it is possible to improve the coding performance, i.e. to improve the quality of the coded image sequence for a given bit rate, or to lower the encoding bit rate for a given quality .

According to an advantageous characteristic of the invention, the method further comprises a preliminary step of determining, based at least on the at least one coding parameter of the current block, said at least two distinct transforms to be applied to said vectors.

The advantage of associating the selection of transforms with the coding parameters is that they adapt to the statistical variations of the block induced by the parameter in question. For example, the encoding parameter considered is the block size, or a prediction mode applied thereto.

For example, preliminary experiments are used to associate coding parameters under consideration with at least two linear transformations. The advantage is that the encoder does not become more complicated.

According to another aspect of the invention, the determining step comprises reading the information stored in the memory, said information comprising at least a coding parameter, an identifier of the first transform, at least one identifier of the first vector of the block or intermediate block, An identifier of at least one second transform different from the first transform, and at least one second vector identifier of the block.

The advantage is that the additional information is not signaled in the bitstream, and the data stored in the memory is replicated on the decoder side.

For example, the memory is organized according to the database. An entry in the database associates the identifiers to be applied to the vector identifiers with a coding parameter transformation.

According to one aspect of the present invention, the transforming step includes a sub-step of rearranging the coefficients of the transformed vectors in the intermediate, individually transformed block.

After application of the linear transformations on the vectors formed in the block to be transformed, the obtained coefficients are rearranged to form a block of coefficients. For example, the coefficients of the first transformed vector are placed on the first line of the block, individually on the first column. This particular embodiment has the advantage that it is simple and has little effect on the encoder.

According to another aspect of the present invention,

Predicting values of the current block from at least one block previously processed according to a mode of prediction selected from a plurality of predetermined modes,

Calculating a residual block by subtracting the predicted values from the original values of the current block

.

According to the present invention, a transform step is applied to the residual current block, and said at least one coding parameter is a prediction mode of the current block.

It has been found that the mode of prediction by itself represents the statistical properties of the residual block, and this relates to associating a particular selection of linear transformations with a particular value of such coding parameters.

According to another aspect of the invention, the method comprises the step of coding said identification of said at least one first transformation and said at least one second transformation.

Information about the linear transformations used is sent in the bit stream. An advantage of this embodiment is that it is suitable for dynamic determinations of transformations by the encoder for each processed block.

According to another aspect, a first transform is applied to a first sub-set of vectors having sizes equal to the size of a line for each of the column sizes of the block, and the at least one second transform is applied to a column of the block Lt; RTI ID = 0.0 > of the < / RTI > vectors of sizes equal to the sizes of the lines of the block, respectively.

For example, two transforms are implemented for each transform sub-step, which are transforms of each of the lines or columns. These two transforms are associated, for example, with the prediction mode of the current block. This embodiment achieves a compromise between compression performance and the cost of storing associations between vector identifiers and conversion identifiers.

According to another aspect of the present invention, the at least one transform sub-step implements a separate transform for each vector of the same size as the size of the line of the block formed in the block, or the size of the column.

An advantage of this embodiment is that it can be precisely adapted to the statistics of each vector of the block to be processed and makes it possible to improve the performance of the encoder in terms of quality and / or compression.

According to another aspect of the invention, the formed vectors belong to a group comprising:

- "Line" vectors of formed pixels of respective coefficients of a line of a current block;

- "Column" vectors of formed pixels of each coefficient of a column of the current block;

- Vectors having a length equal to the length of a line of a block formed by adjacent pixels or coefficients adjacent to the current block from at least two lines of the block;

- Vectors of length equal to the length of a column of blocks formed by adjacent pixels or coefficients adjacent to the current block from at least two columns of the block.

The formed vectors are the same size as the lines or columns of the current block so as not to increase the complexity of the encoder.

The vector according to the present invention is not necessarily exclusively formed with elements of the same line. The non-linear vectors may advantageously be formed from neighboring elements of a block originating from neighboring lines, which may take advantage of certain elements of the considered block, i.e., off-line and columns . This case exists especially for diagonal angle prediction modes, in which the blocks to be encoded have diagonal patterns.

The method just described in its various embodiments may advantageously be implemented by a device for encoding a digital image in accordance with the present invention that includes the following units of a predetermined size that may be implemented for the current block.

- Converting the current block into a transformed block, said block comprising coefficients, said step implementing two consecutive transforming sub-steps, wherein a first sub-step is applied to the current block, Two sub-steps are applied to the intermediate block resulting from the first sub-step, the intermediate block comprising coefficients), and

- Quantizing and encoding the coefficients of the transformed block.

Such a device is special in that it includes a unit that forms at least one first and second separate vectors in a block referred to as the so-called deformable block between the current block and the intermediate block, Wherein the at least one sub-step includes a first transform to the at least one first vector and the at least one sub-step to the at least one sub-step of the block, the pixels being neighboring coefficients of a sequence of length equal to one of the sizes of a block to be transformed, And applying at least one second transform (separate from the first transform) to a second vector of the second vector.

Correspondingly, the invention also relates to a method for decoding a digital image from a bitstream comprising encoded data representing said image, said image being divided into blocks of a plurality of pixels processed in a defined order , The method comprises the following steps implemented for a so-called current block:

- Decoding coefficients of the transformed current block from the data read in the bitstream;

- Dequantizing the decoded coefficients;

- Transforming the current transformed block, wherein the step implements two consecutive inverse transform sub-steps, wherein the first sub-step is applied to the current block and the second sub- ≪ / RTI >

A decoding method according to the present invention comprises:

- The at least one inverse transform sub-step is applied to a block, referred to as the so-called to-be-processed block, from the currently transformed block and the intermediate block, which is at least equal to the length of one line or one column of the block to be processed Applying a first inverse transform to the first vector and applying at least one second inverse transform different from the first inverse transform to at least one second vector of the block of length equal to the length of one line or one column Respectively; And

- Further comprising a sub-step of forming a processed block by positioning the sequences of respective contiguous coefficients that are adjacent pixels of lengths equal to the lengths of each column that is a line from the processed vectors.

According to an aspect of the invention, the method includes a pre-step for determining the at least two distinct transforms to be applied to the first and second vectors of a block to be processed based on at least one coding parameter of a current block, .

According to another aspect, the determining step includes reading in a bitstream of coded data representing at least one first transform and identification information of the at least one second transform.

According to another aspect of the present invention, the determining step further comprises a previous sub-step of forming the first and second vectors in the block to be processed.

According to another aspect of the present invention, a determination step includes reading information stored in a memory, the information including at least a coding parameter, an identifier of a first inverse transformation, at least one identifier of a first vector, Comprises an identifier of at least one second transformation that is distinct and at least one second vector identifier of the block to be processed.

According to another aspect of the present invention, the inverse transform step further includes a sub-step of rearranging the sequences of adjacent coefficients of the block to be processed in the first and second vectors before the at least one sub-step of the transform , And this sequence has a length equal to the size of one of the sizes of the blocks to be processed.

Thus, the decoding method according to the present invention implements inverse operations on the operation of the coding method just described.

Especially:

The sub-steps of the inverse transform correspond to the operations of the transform implemented by the coding method and the inverse operations. They perform each other in reverse order of coding.

The sub-steps for forming a processed block from the sequences of each adjacent pixels that are adjacent coefficients derived from the processed vectors correspond to the inverse operation of the operations forming the vectors to be transformed, which are implemented by the coding method.

The sub-step of rearranging the sequences of adjacent coefficients of the block to be processed in the first and second vectors corresponds to the inverse operation of the operation of rearranging the coefficients of the transformed vectors in the transformed block implemented by the coding method do.

The method just described in the different embodiments is advantageously implemented by a device for decoding a digital image from a bitstream containing encoded data representative of the image, And the device includes the following units that can be implemented for blocks of predetermined sizes, the so-called current block:

A unit for decoding the coefficients of the current block transformed from the data read in the bitstream;

A unit for dequantizing the decoded coefficients;

Unit for inversely transforming the current transformed block, the unit implementing two successive inverse transform sub-units, wherein the first sub-unit is applied to the current block and the second sub-unit is transformed from the first sub- Applied to the resulting intermediate block -;

According to the present invention,

- At least one inverse transform sub-unit of a block, referred to as the block to be processed, from the current transformed block and the intermediate block is transformed into at least a first vector of a length equal to the length of one line or one column of the block to be processed, Apply at least one second inverse transformation different from the first inverse transformation to at least one second vector of the block of length equal to the length of one line or one column of the block to be processed ; And

- Unit in that the inverse transform unit further comprises an inversed sub-unit of the processed block by positioning, from the processed vectors, the sequences of each contiguous pixels that are contiguous coefficients of length equal to the length of each line that is a column .

The invention also relates to a signal carrying a bit stream comprising encoded data of a digital image, the image being divided into blocks of pixels. These signals tell him about the current block:

At least first and second transforms, which are implemented during the coding of the current block during the transforming step of the current block into transformed blocks comprising two consecutive sub-steps of transforming the current block into transform blocks, Applying coded data representative of identification information to a first vector of at least one of the current block and the intermediate block, and applying, to at least a second vector of blocks referred to as a block to be transformed, Applying at least one second transformation;

Coded data representative of at least first and second vectors formed in a block to be transformed from a sequence of adjacent pixels or adjacent coefficients, said sequence having a length equal to one of the sizes of blocks to be transformed It is special in.

The invention also relates to a computer terminal comprising a device for coding a digital image according to the invention and a device for decoding a digital image according to the invention.

The invention also relates to a computer program comprising instructions for implementing the steps of a method for encoding a digital image as described above when the computer program is executed by a processor.

The invention also relates to a computer program comprising instructions for implementing the steps of a method for decoding a digital image as described above when the computer program is executed by a processor.

These programs can use any programming language. They may be downloaded from a communication network and / or recorded on a computer-readable medium.

Finally, the present invention relates to an image processing apparatus, which is readable by a processor and which, in accordance with the present invention, is not integrated or integrated with an encoding device for a digital image and a device for decoding a digital image, A computer program embodying the encoding method, and a computer program embodying the decoding method, respectively.

Other features and advantages of the present invention will become apparent upon reading the following description of one specific embodiment of the invention given by way of example only and non-limiting example with the accompanying drawings:
Figure 1 (already described) schematically illustrates a sequence of digital images to be encoded according to the prior art and the division of these images into blocks;
Figure 2 (already described) shows various probabilities for partitioning a block into sub-blocks according to the prior art;
Figure 3 (already described) schematically shows the steps of a method for encoding a digital image according to the prior art;
4A and 4B (already described) show two examples of approximated frequency transforms according to the prior art;
Figure 5 (already described) shows a comparison table of separable transform complexity measurements and non-separable transforms;
6 schematically shows the steps of a digital image encoding method according to an embodiment of the present invention;
Figure 7A illustrates the elements of the current block;
7B, 7C and 7D illustrate examples of forming "rows" and "columns" vectors from the elements of the current block of FIG. 7A, in accordance with the present invention;
8A shows a first example of estimation of energy values in the pixels of the current block, FIG. 8B shows the uniform regions determined in the block from the estimated energy levels; FIG.
- Figure 9A shows a second example of estimation of energy values in the pixels of the current block, Figure 9B shows the uniform regions determined in the block from the estimated energy levels;
10 illustrates steps for forming vectors and transforming formed vectors according to a second embodiment of the present invention;
Figure 11 shows the compression gains obtained by the encoding method according to this second embodiment compared to the prior art;
12 illustrates steps of forming vectors and transforming formed vectors according to a third embodiment of the present invention;
13 shows compression gains obtained by the encoding method according to this third embodiment compared to the prior art;
Figure 14 schematically shows the steps of a method for decoding a digital image according to an embodiment of the present invention;
15 shows an example of a simplified structure of a coding device for a digital image and a device for decoding a digital image according to an embodiment of the present invention.

The general principles of the present invention rely on the application of separate transforms to different vectors of the same size as the size of the line, each column of the block to be coded.

Consider this an original video consisting of a sequence of M images I1, I2, ... IM, where M is a non-zero integer such as shown in relation to FIG. The images are encoded by the encoder and the encoded data is inserted into a bitstream TB that is sent to the decoder over a communications network or a compressed file FC to be stored on, for example, a hard disk. The decoder extracts the encoded data and then the data is received and decoded by the decoder in a predetermined order known to the encoder and decoder, e.g., time sequence I1, then I2, ..., etc., May vary depending on the embodiment.

With reference now to Fig. 6, now consider the steps of the encoding method according to an embodiment of the present invention.

During the step T0, the block to be processed, the so-called current block x, is selected. For example, this is a CU, square, or rectangular block obtained by partitioning the block CTU.

Below, the sizes of these blocks are height H and width W, which are considered non-zero integers. As an illustrative example, consider a special case where WxH = 4X4 blocks.

Consider the transformation step (T2) of applying the transform to the current block x. The step of transforming the current block T2 is separable and consists of two sub-steps of linear transformation,

- a first sub-step (T21) in which a linear transformation is applied to vectors VI0 to VIH of block x with size W to provide an intermediate block XI comprising WxH coefficients;

- a second sub-step (T24) in which a linear transformation is applied to vectors Vc0 through VcW-1 of block XI with size H to provide a transformed block X,

.

It is noted that according to the invention, the first sub-step can be applied equally well to the vectors Vc0 to VcW-1 of the block x and the second sub-step is applied to the vectors VI0 to VIH-1 of the block XI .

It is understood that different sizes of transformations are applied on the lines and columns to process the rectangular blocks.

According to the invention, at least one of the two transforming sub-steps T21, T24 is implemented from at least two distinct linear transformations, one being formed in the block of pixels x or in the middle block of coefficients XI Is applied to at least one first vector of the same size as the line of the block and the other is applied to at least one second vector of the same size as the line of the block formed in the same block. At the completion of the sub-step (s) using two distinct linear transformations, transformed vectors with the rearranged coefficients at block XI, X of size MxN at T22, T25 are obtained.

In this regard, it should be noted that there are different ways of rearranging the coefficients in the block, and the manner used by the encoder should be known to the decoder. To this end, information representing a rearrangement mode of these coefficients may be encoded and transmitted to the decoder in a bitstream.

Alternatively, predetermined reordering rules may be shared by the encoder and decoder.

Thus, the transform T2 generates a block X that is scanned by the scanning order at T3, quantized at T4, and transformed coefficients ready to be coded at T5. Note that steps T3 and T4 may be interchanged.

At the completion of the processing of the current block, if a predefined coding scanning order is given, it is tested at T6 whether the current block x is the last block to be processed by the encoding unit. If it is the last block, the encoding unit has completed its processing and the encoded data is inserted into the bitstream TB. If it is not the last block, the next step is to select the next block (T0). This block becomes the current block to be processed, and the next step is the prediction step T1 (already described) for determining the transformations to be applied to the current block.

The bit stream TB may then be transmitted to the decoder.

Steps Tl and T2 for determining transitions T1 and T2, in particular sub-steps T20 and T23 for forming vectors, sub-steps T21 < RTI ID = 0.0 & , T24) and sub-steps (T22, T25) of arranging the transformed coefficients into blocks will now be described in detail.

According to a first embodiment of the present invention, said at least two transforms are implemented during a first transform sub-step (T21).

For example, if the first sub-step T21 implements transforms for vectors of the same size as the line size of the current block x and the second sub-step T24 implements transforms for the elements of the intermediate block Xl And implement at least one transformation on vectors of the same size as the size of the column.

In this case, the method implements a preliminary step T20 to form H vectors (Vl0-VlH-1) of length W from the pixels of the current block (x).

Advantageously, these vectors are formed such that each element of the current block is used in a single vector.

With reference to Figure 7A, we consider the current block (x) of size 4x4 as an example. It contains 16 coefficients (x0 to x15).

7B, a first example of vectors VlO to Vl3 formed from the lines of block x is shown. This type of vector (Vlh) corresponds to the line number (h) of the current block (x).

7D, a second example of vectors (V'i0 to V'11) formed from four elements neighboring between pairs of block (x) is shown. These items do not originate from the same line. For example, the vector (Vl'0) includes three consecutive elements (x0, x1, x2) of the first line of the block adjacent to the element (x0) of the first line and one Element x4.

It is appreciated that this type of vector of the same size as the size of the line can be advantageously used to monitor the texture discontinuities present in the block and better utilize the correlation between the elements of the vector.

The coding method according to the invention then comprises at least two separate linear transforms (L0, L1) to be applied to the formed vectors (V10 to HIV-1), at least one first transform L0) and at least one second transform (L1) to be applied to at least another vector (Vlh2r), wherein h1, h2 are integers between 0 and H-1, and h1 ≠ h2.

The transformations to be distinguished may be a DCT or DST type or any other effective linear transform for encoding. Thus, optimal transforms can be used for correlation, i.e., KLT ("Karhunen-Loeve Transform"), or "Robust learning of 2D separable transforms for next generation" published in the conference DCC quot; video coding "by Sezer et al.

Advantageously, the at least two linear transformations are determined based on at least one encoding parameter of the current block, for example a block size or also a selected intra prediction mode.

For example, previously conducted offline experiments identified at least two linear transformations adapted to a particular intra prediction mode and stored the acquired associations in memory.

Advantageously, the identifiers of the linear transformations associated with the coding parameter values are stored in the coder memory. This may be accomplished, for example, by means of entries relating the encoding parameters, such as, for example, the previously mentioned intra prediction mode, the identifier of the first transformation, the identifier or index vector of the intermediate block or block to which the first transformation is to be applied, The identifier of the transform, the identity of the transform, the identity of the at least one identifier of the block to which the second transform is applied or the index of the second vector, and the identifier of the at least one second transform.

8A, from the prediction mode number 26 according to the standard intra-prediction mode, in this case, the standard HEVC, average energy values per pixel in the remaining blocks of size 8x8 were estimated. Such an estimate is made offline.

It is noted that the energy per pixel has a horizontal stripe pattern that justifies cutting into two (or more) distinct regions.

With reference to Figure 8b, two regions R1 and R2 are shown:

The first region R1 is composed of the first three lines with a certain pattern;

The second region R2 consists of the last five lines with a less constant profile due to the more marked discontinuities on the first column.

Advantageously, two transforms L1 and L2 have been determined, one applied to the line vectors of region R1 and the other applied to the line vectors of region R2.

Thus, in accordance with this prediction mode, the encoder applies a partition to two " line " transforms each sharing one region.

For an apparently homogeneous region (R1), a transformation or DCT-type transformation defined for KLT was chosen. Since the DCT is suitable for transforming successive patterns, this may be considered suitable.

For region R2, a transformation is selected that is capable of taking into account a larger discontinuity in the first column, e.g., DST, or KLT. Since DCT is suitable for transforming patterns with initial discontinuities, this can be considered suitable. It is noted that for region R2, the average energy of the first pixel on the left edge has a value that is significantly different from the values of the other pixel energies.

Without memory constraints, possibly as many more areas as the lines will be defined, and eight conversions will be used at the expense of a larger memory footprint.

Taking memory occupancy constraints leads to reduced efficiency and limited performance reduction.

A transformation L0 for the line vectors vl0 to vl2 and a transformation L1 for the column vectors vl3 to vl7, in accordance with the example of Figures 8A and 8B.

With respect to FIG. 9A, a given coding mode, in this case, is obtained for the residual signal obtained by offline pre-analysis of blocks of size 8x8 from the prediction mode 19 defined by the HEVC standard, Energy is now provided. This prediction mode has an angle of approximately -53 degrees and performs diagonal prediction from top to bottom.

The energy variations between the pixels illustrate the three zones R'1, R'2, R'3 shown in FIG. 9B. They must be cut to process them by three separate transformations associated with the regions (R'1, R'2 and R'3) and adapted to their respective statistics.

In this example, it is seen that the regions formed do not correspond to one or more lines of the block. In accordance with the present invention, from neighboring elements, it is of interest to form vectors of sizes equal to the size of the line, while all of them do not necessarily belong to the same line depending on the boundaries of the considered regions. For example, the vector Vlh1 shown in Fig. 9A corresponds to the region R'1.

According to a first aspect, the input of the database BD1 comprises a prediction mode 19, an identifier of the transform L'0, an identifier of the vector (s) formed in the region R'1 to which the transform L'0 should be applied, 1, the identifier of the vector (s) formed in region R'2 to which transform L'2 should apply, the identifier of transform L'3, the vector (s) formed in region R'3, Lt; / RTI >

Accordingly, the coding method determines how to form vectors of sizes equal to the sizes of one line by reading the corresponding entries in the database BD1, and based on the used prediction mode, Lt; / RTI >

According to another aspect, the coding method is based on the prediction mode INTRA number 19 and the pattern areas R'1, R'2, R'3, which were presented immediately before, at the end of prediction, the correlation between the values of the pixel residuals for the current block Lt; / RTI >

In the case of a high correlation between the two quantities, i. E. In the case of a correlation on a set correlation threshold, the encoder decides to use three transforms L'0, L'1, L'2.

The encoder applies the three transforms adapted for the different zones together with the transformations adapted to the zones. For example, adaptation in terms of KLT is performed.

It indicates this selection by inserting a decoder into the bitstream coded information representing the pattern indicator associated with the INTRA prediction mode number 19.

According to this embodiment, it is understood that the database BDl potentially includes a plurality of entries corresponding to the same encoding parameter, and each entry includes a distinct pattern indicator associated with different vectors and different transitions.

Upon receiving the prediction mode information and the pattern selected by the encoder, the decoder reads in its database (BD2) the identifiers of the vectors to be formed in the block and the identifiers of the transformations to be applied to the vectors. The decoder then performs the inversions of the transformations performed in the encoder.

Alternatively, the linear transforms are determined dynamically by pre-analysis of the current block to be encoded. For example, such pre-analysis implements known outline analysis techniques using estimates of the gradient. The contour detection is then utilized to determine the at least two regions and assign the transformation types to the regions, depending on the characteristics of the at least two regions, e.g., the texture homogeneity of the regions. In this case, the identifiers of the determined linear transformations and the identifiers of the associated vectors are signaled in the bitstream and transmitted to the decoder.

Alternatively, such outline analysis may be performed on one or more neighboring blocks that are already processed and combined with the continuity assumptions for the current block, e.g., according to the outline orientation in the next block associated with the current block .

If the contour continuity determination is made, then the regions of the current block are then determined based on the regions of the neighboring block that have already been processed.

In this case, the encoder signals the decoders based on the coded information representative of the inheritance mode for the affected neighboring block, i.e., the neighboring blocks into which the vectors to be used and transforms should be inherited. It should be appreciated that once a decoder has been decoded, it is required to implement the same outline analysis for the same neighboring block to deduce vectors, that is, transformations to be used, from the analysis.

According to an embodiment variation of the present invention, the at least two distinct linear transformations are implemented during the second transform sub-step T31.

In this case, vectors of size H equal to the size of the columns of the current block are formed from the coefficients of the intermediate block XI at (T11) following the first transform sub-step T30. W vectors Vc0 through VcW-1 of length H are obtained.

As above, several types of vectors of size H can be formed.

With reference to Figure 7C, the vector VCW corresponds to the column number w of the block.

With reference to Figure 7E, the vector Vc'w includes elements from two adjacent columns.

The determining step T'2 provides at least two distinct linear transforms C0, C1 to be applied to the formed vectors Vc0 through VcW-1.

Various embodiments of the steps of forming the vectors VI0 through VIH-1 and determining the linear transformations Li presented above may be transposed into steps of forming the vectors Vc0 through VcW-1 and determining the linear transformations Cj, Note that they are implemented for intermediate block XI.

Advantageously, the principles of the present invention are implemented in two transform sub-steps T30 and T31. In this case, the encoding method comprises a first step (T10) of forming vectors VI0 through VIH-1 of size W from the current block X, transforms L0, L1 to be applied to vectors VI0 through VIH-1 A second step T11 for forming vectors Vc0 through VcW-1 of size H from the intermediate block XI and a second step T11 for determining transforms C0 and C1 to be applied to the vectors Vc0 through VcW- The second step (T21) is implemented.

One advantage is to better utilize the statistical features of the different vectors formed for the two transforming sub-steps. Of course, on the other hand, this embodiment requires storing a greater number of transforms.

Referring now to Fig. 10, a method for encoding a digital image according to a second embodiment of the present invention is now illustrated.

According to this embodiment, the present invention considers H-line transforms L0 to LH-1 and W column transforms C0 to CW-1. In other words, regardless of whether the vectors Vl0 to VlH-1 of size W were formed at (T'20) or whether the vectors Vc0 to VcW-1 of size H were formed at (T'23) The vector - specific linear transformation is applied to the current block.

As an illustrative example, consider the case of a 4x4 block.

The converted signal X1 during step T'21 is obtained by concatenation of the following operations:

L ₀ , L ₁ , L ₂ and L ₃ represent line transforms, and thus in this embodiment, are a matrix of size 4x4, which is a DCT / DST-type, As a high-speed algorithm.

Thus, the 16 transformed coefficients X10 ... X15 can be obtained from the 16 (x0 ... x15) pixels of the initial block x. They are rearranged at T'22 to form a block Xl, for example, by considering that each vector Vli converted by the linear transformation Li helped form the line i of the block Xl.

The column transformation is similarly applied to T'24 as follows:

C ₀ , C ₁ , C _2, and C ₃ represent column transforms, and thus are matrices of size 4 × 4 in this embodiment, which is especially useful when the transformations used are DCT / DST- Based on the same butterfly decomposition, it can potentially be implemented as a fast algorithm.

Thus, in the embodiment of block 4x4 which was just above, eight transforms were provided.

The advantage of this embodiment is that it takes into account the fact that each line and each column individually provides their own statistics. Compression performance is improved.

Referring to Fig. 11, for a block of size 4x4, the gain provided by using multiple transforms in the specific case of INTRA coding in HEVC is shown.

In this context, the HEVC uses line and column transforms of type DST VII for blocks of size 4x4, which showed the best results in signal miniaturization.

This occurs in the context of intra prediction number 18. This prediction is diagonal. To evaluate the performance of the transform, we measure its coding gain, that is, the ratio between the energy of the signal and the geometric mean of the transformed coefficients.

With reference to the table of FIG. 11, we have shown that the DSTs used in the HEVC, the DCTs commonly used in image / video encoding, the optimal separable conversion in the sense of KLT (designated Sep in the table) , And the set of transforms (designated as mSEP in the table) according to the present invention.

In the sense of KLT, optimization is to find the lines and columns transformations that provide the best coding gain. According to the prior art, KLT transforms are obtained by considering the pixel-to-pixel correlations of the vectors to be transformed and at most by determining a transform that de-correlates these pixels. Thus, the autocorrelation matrix of the pixel is determined, and then diagonalization is performed: The appropriate vectors to generate the correlation release form the transform KLT.

These results demonstrate improved performance using the present invention or considerable advantages over prior art methods.

Referring now to Fig. 12, a method for encoding a digital image according to a third embodiment of the invention is now illustrated.

According to the present embodiment, from the block of pixels x is formed "line" vectors at T''20, and two separate line transforms are applied to them during the first sub-step T''21. The obtained coefficients are rearranged to T''22 to form intermediate block Xl, while the "column" vectors are formed at T''23 and two distinct column transforms are formed during the second transform sub- The transformed coefficients are rearranged at T''25 to form the transformed block X. [

Thus, the line transform implemented by the first sub-step T''21 distinguishes the transform for the first line from that used for others, and for a block of size 4x4 it can be expressed as:

So we now have only two transforms (L0 and L1), not the previous four.

In the case of columns, it can proceed in a similar fashion, i.e., only keeps two transforms to distinguish columns (e.g., the first of the following columns). It is also possible in one embodiment to maintain only one transformation for a set of columns, or in another embodiment, to retain four distinct transformations as outlined in the first embodiment.

Thus, the impact associated with storing transforms is significantly reduced.

To illustrate the performance obtained by the third embodiment for memory reduction, the coding gains achieved by blocks 4x4 are again illustrated for prediction 18 of the intra mode of the HEVC.

Two separate transforms are maintained for the lines and columns. Thus, we have a total of four transforms instead of eight as shown in the first embodiment (note configuration 4Sep).

Alternatively, two separate transforms are maintained for the lines and only one transformation is maintained for the columns (in the case of the columns, we are close to the state of the art). Thus, we have a total of three transforms instead of 8 as shown in the first embodiment (note configuration 3Sep).

The results are shown in the table of FIG.

It is therefore evident that these versions with reduced memory can be closer to the results obtained by the version with 8 transforms (mSep) in half or less of the memory used. There are significant advantages in terms of compression with regard to the latest technology transformations.

With reference to FIG. 14, we now consider the steps of a decoding method according to an embodiment of the present invention.

We assume that the bitstream (TB) has been received by a decoding device that implements the decoding method according to the present invention.

At D0, we first select the first block to be processed as the current block C '. For example, this is the first block (in lexicographic order). This block contains MxN pixels, where M and N are non-zero integers.

As discussed for the encoding process, the considered block C 'is the sub-block CU obtained by cutting the block CTU or block CTU, or the remainder-block CU obtained by subtracting the block or current block from the current block, Block.

In step D1, the coded data on the current block C 'is read and decoded. At D2, the decoded data for the current block C 'is dequantized. We obtain the vector DQ [k] of dequantized coefficients, where k is an integer from 0 to MxN-1.

During step D3, the coefficients of the vector DQ are arranged in the transformed block X '. This is the inverse operation of the course T3 implemented by the encoding methods.

At D4, the transforms to be applied to the current block X 'are determined during successive sub-steps D51 and D54 of the inverse transformation. According to the invention, one of them implements at least two distinct linear transformations. These transforms will be understood to be the reverse of those used by the coding method according to the present invention.

In the decoder, the inverse transforms to be applied to the vectors of the same size as the respective lines of the columns may be determined in various manners, of which the following may be mentioned as examples:

- By reading a bit stream (TB) of at least one first transform and one second transform in the identification information;

- By reading the identification information of the transformations used by the encoder in local or remote memory;

Advantageously, the selection of the first and second transforms may be associated with an encoding parameter of the current block, e.g., its size or its prediction mode.

According to one embodiment of the present invention, the transforms to be applied include a memory, for example an entry of a database BD2 associated with an encoding parameter, an identifier of a first transform for at least one identifier of a first vector of a block or intermediate block, Is obtained by reading the identifier of at least one second transformation (separate from the first transformation) for at least one identifier of the second vector of the first vector of the second vector.

Advantageously during D4, we also determined how vectors were formed in the current block in the encoder. For example, the vector identifier corresponds to a vector of known type formed in a block of size MxN.

It will be appreciated that the decoder needs to know how to form the vectors in the same way as the encoder. Several types of column or line vectors may be formed. Similar to what we have described for the various transforms to be applied, several embodiments are contemplated including:

- Decoders have the same rules as encoders for locally forming vectors. For example, these rules are stored in memory;

- Advantageously, the decoder accesses information stored in a database that includes entries associated with the encoding parameters, which are indications that allow the formation of a W vector of size (H) and an H vector of size (W) in the current block. This database is replicated at the level of the encoder and decoder.

- The decoder extracts information representative of the type of vector formed by the encoder from the bitstream. This variation is particularly advantageous when the type of vector being formed is dynamically selected based on pre-analysis of the contents of the current block implemented at the level of the encoder.

At the completion of this step, we then know the number, indices and types of vectors associated with the linear transform identifier.

In the following description, during encoding, the first sub-step implements at least two transforms of vectors of the same size as the line, and the second sub-step implements one or more transforms .

Upon decoding, the inverse operations will be performed in reverse order to that performed during encoding. In this exemplary embodiment, the first sub-step D51 of the inverse transformation is applied to vectors of the same size as that of the column accordingly.

During step D50, vectors having coefficients of the same length as those of one column of block X 'are formed in block X'. This is the inverse operation of the array operation (T25) of the coefficients of the transformed vectors in block X implemented by the encoding method.

For example, on the encoder side, the coefficients of the column vector Vcj transformed by the linear transformation C0, C1 or Cj of step T24 are simply positioned in the column j of the block X. In step D50 of the decoding method, conversely, a vector V'cj of the same length as the column is formed from the coefficients of column j of block X '. Thus, in this example, it is understood that the column of X 'includes coefficients from a linear transformation applied to the associated vector (vcj) by the coding method. In this simple case, step D50 is then simply configured to form the "column" vectors V'cj from the columns of block X '.

Advantageously, step D50 is based on information about this rearrangement obtained during decision step D4. For example, the reordering, which relates column cj0 in the entry of the linear transformation (C0- ¹ ) and column (cj1) in the entry of the linear transformation (Cl- ¹ ) Or associates vector identifiers with respective linear transforms.

These vector identifiers advantageously include one type of vector and a vector index. For example, the type of vector considered is the column type and the vector to be formed is number (j), which corresponds to column (j) of the block.

In D51, linear inverse transform ^(-1 C0, C1 ^-1) is the vector formed from D50 (V'cj0, V'cj1), for example, when they are perpendicular or use the fast algorithm of the formed substitution column (cj0, cj).

Vectors v'cj0, v'cj1 are obtained.

During step D52, the coefficients of the processed vectors v'cj0, v'cj1 are positioned to form the intermediate block X'1. To this end, we use identification information of the vectors relative to the formation of the vectors on the encoder side and determined in step D4, before proceeding to the opposed operation.

For example, if the vectors are column vectors, the identification information includes a column index and the elements of the vectors are arranged in corresponding columns.

If the vectors are not linear, identification information identifying the type of nonlinear vector is known to the decoder, which can position the elements in block MxN. For example, the information identifying that the vector vcj formed on block Xl on the encoder side is of the type described with respect to Figure 7e, with an index corresponding to vector vc0, If indicated, the coefficients will be placed at X'1, X'0, X'4, X'8.

The resulting intermediate block X'1 is implemented in the subsequent forming step D53 and the forming step is similar to the step D50 before passing through the second sub-step of the inverse transformation D54, forming a vector (V'li) in the block (X'l) from the obtained information by rearranging, and associates this end, in one of the linear transformation also determined at D4 ^(-1 LO and L1 ^-1).

In D54, the inverse transforms are applied in a permuted form, if they are orthogonal or using fast algorithms, to produce vectors (V'li0, V'li1) of the same size as the line. Operations implemented by the decoder according to the present invention are similar to those used by the state of the art: therefore, complexity is not changed to perform transformations.

At D55, the obtained vectors are positioned to form a block (x ') of pixels. This is the inverse operation of forming T20 vectors implemented in the coding method according to the present invention.

At D6, a block of pixels of the decoded image from the obtained block (x ') is reconstructed and the block is integrated with the ID picture to be decoded. If block (x ') is the remaining block, a prediction of the current block obtained from the pre-processed reference image will be added to the block.

During step D7, if a predefined route order is provided, it is tested whether the current block is the last block to be processed by the decoding unit. If so, the decoding process completes its processing. If not, the next step is to select the next block D0 and the decoding steps D1 to D7 described above for the next selected block are repeated.

It will be noted that the invention just described can be implemented using software and / or hardware components. In this context, the terms "module" and "entity " used in this document may be a software component or a hardware component or even a set of hardware and / or software capable of implementing the described functions for the associated module or entity .

Referring now to Fig. 15, an example of a simplified structure of a digital image encoding device 100 according to the present invention is now presented. The device 100 implements the encoding method according to the present invention as described in connection with Fig.

For example, the processing unit 110 that is driven by the device 100 is a computer program (Pg ₁₎ (120) for implementing the method according to the invention is stored in a processor (μ1) is mounted, and the memory 130 .

At initialization, computer program code instructions (Pg ₁ ) 120 are loaded into RAM, for example, before being executed by the processor of the processing unit 110. The processor of the processing unit 110 implements the above-described method steps in accordance with the instructions of the computer program 120. [

In this embodiment of the invention, the device 100 includes a first sub-unit TR1 for converting the current block into an intermediate block and a second sub-unit TR2 for converting the intermediate block to a transform block At least one unit (TRANS) for transforming the current block into a transformed block (X), a unit (QUANT) for quantizing the transformed block, a unit (COD) for coding the quantized block, (INSERT) for inserting into the bit stream (TB).

In accordance with the present invention, the transform is a transform between at least one of the transforming sub-units and a sub-unit (ARR) for arranging the coefficients obtained in the block, And at least one sub-unit FORM for forming at least two vectors from one element (pixels, coefficients each). The device also includes a unit (DET) for determining at least two distinct transforms to be applied to the vectors based on at least the coding parameters of the current block.

Advantageously, the device 100 includes an encoding parameter and an identifier of a first transform having at least one identifier of a vector of blocks or intermediate blocks, a first transform having an identifier of at least one of the second vector of the block, For example, a unit (BD1) for storing a table including entries associating identifiers of at least one second conversion.

These units are controlled by the processor 1 of the processing unit 110. [

Advantageously, such a device 100 may be integrated into a user terminal (TU). Next, the device 100 is arranged to cooperate with at least the following modules of the terminal TU:

- An E / R module for transmitting and receiving data to be transmitted to the decoding device via a telecommunication network with a bit stream (TB) or a compressed file (FC);

For example, the decoding device 200 may be coupled to a computer program Pg2 220 that implements the decoding method according to the present invention as described above with reference to FIG. 14, (Not shown).

At initialization, the computer program code instructions (Pg2) 220 are loaded into the RAM, for example, before being executed by the processor of the processing unit 210. [ The processor of the processing unit 210 implements the above-described method steps in accordance with the instructions of the computer program 220. [

In this embodiment of the invention, the device 200 comprises at least one unit (DEC) for decoding the coefficients of the current block transformed from the data read in the bitstream, a unit for dequantizing the decoded coefficients (DEQUANT), 2 consecutive inverse sub-unit comprises a (TRAN ^-1) for the inverse transform, converting the current block is adapted to implement unit and the first sub-unit (TR1 ^-1) is the current transformation block is applied to the second sub-unit (TR2 ^-1) is the first sub-block is applied to the intermediate arising from the unit. According to the present invention, the sub-units ^(-1 TR1, TR2 ^-1) of at least one implement, individual at least a first and a second linear transformation to each other by so-called the processing for a block to be converted from the current block and the intermediate block do.

According to the present invention, the inverse transform unit further comprises a sub-unit (FORM- ¹ ) adapted to rearrange coefficients of the vectors processed by the first and second transformations of the processed block.

Advantageously, the device further comprises a unit (ARR-1) for forming at least two vectors from the coefficients of the block to be processed, said vectors having a length equal to one of the sizes of the current block to which the first and second linear transformations are to be applied And a unit (DET) for determining at least two distinct linear transformations to be applied to the vectors based on at least the coding parameters of the current block.

These units are controlled by the processor 2 of the processing unit 210. [

Advantageously, such a device 200 may be integrated into a user terminal (TU). Next, the device 200 is arranged to cooperate with at least the following modules of the terminal TU:

- A module (E / R) for transmitting and receiving data received from the telecommunication network by a bit stream (TB) or a compressed file (FC);

- A device (DISP), e.g., a terminal monitor, for generating images that present a decoded digital image or a series of decoded images to a user.

Advantageously, the user terminal (TU) may comprise both an encoding device and a decoding device according to the invention.

It is needless to say that the embodiments described above are given as purely illustrative and non-limiting examples, and that many modifications can be readily made by those skilled in the art without departing from the scope of the present invention.

Claims (17)

CLAIMS 1. A method for encoding a digital image,
The image is divided into a plurality of blocks of pixels processed in a predetermined order,
The method includes steps that are implemented for a current block of predetermined sizes,
The steps,
- transforming the current block into a transformed block (T2), the block comprising coefficients, the step implementing two consecutive transforming sub-steps, wherein the first sub-step (T21) Block and a second sub-step (T24) is applied to an intermediate block resulting from the first sub-step, the intermediate block comprising coefficients, and
- quantizing (T4) and encoding (T5) the coefficients of the transformed block,
Wherein the transforming step is performed prior to at least one transforming sub-step among the transforming sub-steps, wherein the sub-step is applied to a block that is a so-called transformed block among the current block and the intermediate block, Further comprising spare sub-steps (T20, T23) for forming at least first and second discrete vectors in said vector, said vector being a coefficient of a sequence of pixels or of a pixel, The adjacent coefficients of the block to be transformed, which are the same length as one; And
- said at least one transform sub-step comprises applying a first transform (L0, C0) to said at least one first vector and applying to said at least one second vector of said block at least one 2 transforms (L1, C1).
A method for encoding a digital image. The method according to claim 1,
Further comprising a preliminary step (T1) for determining, based at least on the coding parameters of the current block, at least two distinct transforms to be applied to the vectors,
A method for encoding a digital image. 3. The method according to claim 1 or 2,
The determining step (T2) comprises reading the information stored in the memory,
Wherein the information comprises at least one of a coding parameter, an identifier of the first transform, at least one identifier of the first vector of the block or the intermediate block, at least one identifier different from the first and at least one second vector identifier of the block, Lt; RTI ID = 0.0 > a < / RTI >
A method for encoding a digital image. 4. The method according to any one of claims 1 to 3,
Wherein the transforming comprises sub-steps (T22, T25) for rearranging the coefficients of the transformed vectors in the intermediate block (XI), individually the transformed block (X)
A method for encoding a digital image. 5. The method according to any one of claims 1 to 4,
- predicting values of the current block from at least one block previously processed according to a prediction mode selected from among a plurality of predetermined modes,
Calculating residual blocks by subtracting the predicted values from the original values of the current block,
Wherein the transforming step (T2) is applied to a residual current block, and the at least one coding parameter is a prediction mode of the current block,
A method for encoding a digital image. 6. The method according to any one of claims 1 to 5,
The identification information coding step of the at least one first transformation and the at least one second transformation,
A method for encoding a digital image. 7. The method according to any one of claims 1 to 6,
Wherein the first transformation is applied to a first sub-set of vectors equal in size to the line, and individually of a size equal to the size of the column of the block, and wherein the at least one second transformation comprises: The same applies to a second sub-set of vectors of the same size, individually the same size as the column of the block,
A method for encoding a digital image. 7. The method according to any one of claims 1 to 6,
Wherein the at least one transform sub-step implements separate transforms for each vector of the same size as the size of the line of the block,
A method for encoding a digital image. A device (100) for encoding a digital image,
The image lk is divided into a plurality of blocks of pixels processed in a predetermined order,
The device includes units that can be implemented for a current block (x) of predetermined sizes,
The units,
A unit (TRANS) for converting the current block into a transformed block, the block comprising coefficients, the unit implementing two successive transform sub-units, the first sub-unit (TR1) Unit and a second sub-unit TR2 is applied to an intermediate block originating from said first sub-unit, said intermediate block comprising coefficients, and
- a unit for quantizing (QUANT) and encoding (COD) the coefficients of the transformed block,
- said converting unit further comprises a sub-unit (FORM) for forming at least first and second separate vectors in said current block and said block being a so-called block to be transformed from said intermediate block, , The coefficients of the sequence of adjacent pixels individually, and the adjacent coefficients of the transform block, of the same length as one of the sizes of the blocks to be transformed individually, and said sub-units (TR1 , TR2); < / RTI > And
The at least one transform sub-unit applies a first transform to the at least one first vector and at least one second transform separate from the first transform to the at least one second vector of the block Including,
A device (100) for encoding a digital image. CLAIMS 1. A method for decoding a digital image from a bit stream (TB) comprising encoded data representing a digital image,
The image lk is divided into a plurality of blocks of pixels processed in a predetermined order,
The method includes steps that are implemented for a block that is a so-called transformed block (X '),
The steps,
- decoding (D1) the coefficients of the current transformed block from the data read in the bitstream;
- dequantizing the decoded coefficients (D2);
(D5), which inversely transforms the current transformed block, wherein the step implements two consecutive inverse transform sub-steps, wherein a first sub-step (D51) is performed on the current block and a second sub-step ) Applied to the intermediate block resulting from the first sub-step,
- the at least one inverse transform sub-step is applied to a block which is a so-called to-be-processed block from the currently transformed block (X ') and the intermediate block (X'1) (C0 ^-1 , L0 ^-1 ) to at least one first vector (V'cj0, V'li0) of the same length as the length of one line of one column or one column At least one second inverse transform (C1 ^-1 , L1 ^-1 ) different from the first inverse transform is applied to at least one second vector (V'cj1, V'li1) of the block having a length equal to the length of the column, ≪ / RTI > And
- said inverse transforming step comprises the steps of: extracting from the processed vectors (v'cj0, v'cj1, v'li0, v'li1) a column, individually adjacent pixels of length equal to the length of the line, Further comprising sub-steps (D52, D55) for forming the processed block (X'1, x ') by positioning the sequences.
A method for decoding a digital image. 11. The method of claim 10,
At least on the basis of the coding parameter of the current block, at least two separate transformation applied to the vector pre-step for determining the ^(-1 C0, C1 ^{-1, -1} L0, L1 ^-1) (D4) Further included,
A method for decoding a digital image. The method according to claim 10 or 11,
Wherein said inverse transforming step comprises prior to said at least one transforming sub-step, a sequence of adjacent coefficients of a block to be processed in said first vector and said second vector (V'cj0, V'cj1, V'li0, V'li1) Further comprising sub-steps (D50, D53)
The sequence having a length equal to one of the sizes of the blocks to be processed,
A method for decoding a digital image. A device (200) for decoding said digital image from a bit stream (TB) comprising encoded data representing a digital image,
The image is divided into a plurality of blocks of pixels processed in a predetermined order,
The device includes units that can be implemented for blocks of predetermined sizes which are so-called current blocks,
The units,
A unit (DECOD) for decoding the coefficients of the current transformed block from the data read in the bitstream;
A unit (DEQUANT) for dequantizing the decoded coefficients;
- a unit for inversely transforming the currently transformed block (TRANS ^-1 ) - the unit may implement two consecutive inverse transform sub-units, the first sub-unit for the current block and the second sub- Applied to an intermediate block originating from the first sub-unit,
- the at least one inverse transform sub-unit of blocks (X ', Xl) being the so-called to-be-processed block from the currently transformed block and the intermediate block, respectively, of the columns of the block to be processed, (C0 ^-1 , L0 ^-1 ) to at least one first vector (V'cj0, V'li0) of the coefficients of the block to be processed of the same length, and a first inverse (V'cj1, V'li1) of at least one of the coefficients of the block to be processed, the length of which is equal to the length of the line, and at least one second Applying an inverse transform (C1- ^l , L1- ^l ); And
- said inverse transform unit D5 comprises a column from the processed vectors (v'cj0, v'cj1, v'li0, v'li1), each adjacent pixel of length equal to the length of the line individually, Further comprising a sub-unit (D52, D55) forming a processed block (X'1, x ') by positioning the sequences of contiguous coefficients.
A device (200) for decoding a digital image. As a signal carrying a bit stream (TB) comprising coded data of a digital image,
The image is divided into blocks of pixels, and for the current block,
- coded data representing at least one first transformation and at least one second transformation identification that are distinct during coding of the current block during the step of transforming the current block into a transformed block (T2) Wherein the step (T2) implements two consecutive sub-steps (T21, T24) for transforming the current block into a transformed block, the at least one transforming sub-step comprising at least one first vector (vlh1, vcw1) of a block that is a so-called block to be transformed among the current block and the intermediate block (x, Xl) to the first vector (vlh1, vcwl) And applying at least one second transformation (L1, C1) that is distinct from the first transformation
- coded data representing at least the first vector and the second vector formed of the block to be transformed, the vector having a length equal to one of the sizes of the block to be transformed, The coefficients, individually including a sequence of adjacent pixels,
signal. As a user terminal (UT)
A device (100) for encoding a digital image according to claim 9 and a device (200) for decoding a digital image according to claim 13,
A user terminal (UT). As the computer program Pg1,
Comprising instructions for implementing a method for encoding a digital image according to any one of claims 1 to 8 when executed by a processor,
Computer program (Pg1). As the computer program Pg2,
13. A computer-readable medium having stored thereon instructions for implementing a method for decoding a digital image according to any one of claims 10 to 12, when executed by a processor,
Computer program (Pg1).

Images